Was set defensively without knowing the actual GPU arch; if ROCm
supports the card natively, the override is at best a no-op and at
worst masks the real arch. Add it back with the right value if the
service actually fails to detect the GPU.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Runs llama.cpp's ROCm build under DynamicUser, with the HF model cache
in StateDirectory (survives systemctl clean) and KV slot saves in
CacheDirectory. Listens on :8000.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The full nix-ld library list shadowed nix's own curl, breaking
libnixstore.so with "CURL_OPENSSL_4 not found". The prebuilt node
watcher binding only needs libstdc++/libgcc_s, so use stdenv.cc.cc.lib
and let nix-built tools resolve their own deps via RUNPATH.
The file watcher binding (and other node-precompiled .node modules
loaded via dlopen) failed with "libstdc++.so.6: cannot open shared
object file" because systemd services don't inherit the user shell's
LD path. Reuse the nix-ld library list so the service sees the same
common libraries unwrapped binaries get globally.
The opencode-serve unit ran with systemd's minimal default PATH, so
shell commands invoked by the agent (git, make, nix, node, rg, etc.)
were not found. Set systemd.services.opencode-serve.path on both sgx
and amd to a common dev toolset.
Mirror of the sgx opencode setup: systemd service on port 4196 fronted
by nginx with a per-host ACME cert (DNS-01 via internetbs). Adds amd
key + path rule to .sops.yaml so secrets under .secrets/amd/ encrypt
for the host.
Wire up restartUnits on secrets whose consumers cache them in memory
(daemons read at startup), so sops-nix restarts the affected unit on
activation when the decrypted content changes:
- firefly: app_key → phpfpm-firefly-iii;
auto_import_secret + access_token → phpfpm-firefly-iii-data-importer
- searx: secret_key → uwsgi
- opencode: web password → opencode-serve
- mail: sasl_passwd → postfix
- forgejo: gitea_dbpass → forgejo; runner-token → gitea-runner-default
Secrets read on demand by oneshots/timers (firefly sparda_pin, ntfy
token, restic backup creds, acme dns creds, wg conf) are left as-is.
Adds metacfg.cli-apps.opencode (default enabled) which mounts the
in-repo opencode config (provider list, web-search skill) via
xdg.configFile, so all hosts pick it up automatically.
Bulk imports of 100+ transactions per chunk hit the default 60s
fastcgi timeout on the main Firefly III vhost too — not just the
importer endpoint. The importer's per-transaction API call to Firefly's
/api/v1/transactions can take 20+s on a fresh DB without ANALYZE,
which compounds with the 30s PHP max_execution_time cap.
- nginx fastcgi_read_timeout=600s on both `firefly` and `firefly-import`
vhosts
- php_admin_value[max_execution_time]=600 + memory_limit=512M on both
PHP-FPM pools
- VANITY_URL on the importer now points to the main Firefly III URL
(was wrongly pointing at the importer's own domain, breaking
clickable transaction-show links in importer log messages)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
SQLite was slow under btrfs CoW, and the no-CoW migration path turned
out to be fragile (WAL deletion without checkpoint = data loss). Move
to PostgreSQL on Unix-socket peer auth — no password needed for the
local-host setup, NixOS provisions the database+user declaratively.
Drop the now-unused +C tmpfiles rule on the sqlite directory; the
leftover database.sqlite* files at /var/lib/firefly-iii/storage/database/
are harmless and can be removed manually after switch is verified.
Migration of existing Firefly III data is not preserved by this
commit — fresh-start path: re-register admin, re-issue PAT, re-POST
the bulk CSV through the importer.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Random-write SQLite traffic fragments CoW filesystems quickly. The `h`
tmpfiles directive sets +C on the database directory; new SQLite files
(WAL, SHM, recreated main DB) inherit no-CoW automatically. No-op on
non-btrfs filesystems.
Migration of existing files must be done manually with checkpoint-first:
systemctl stop phpfpm-firefly-iii.service
sqlite3 .../database.sqlite 'PRAGMA wal_checkpoint(TRUNCATE);'
# then recreate main file inside the +C dir
systemctl start phpfpm-firefly-iii.service
Skipping the wal_checkpoint and naively deleting .sqlite-wal will lose
all writes that haven't been checkpointed (PHP-FPM SIGTERM does not
trigger a checkpoint).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
End-to-end verified: aqbanking-cli fetches Sparda Südwest transactions
via FinTS PIN/TAN + SecureGo+, exports CSV using a custom decimal-amount
profile, POSTs to firefly-iii-data-importer's autoupload endpoint, which
creates transactions in Firefly III via API.
Changes vs. previous WIP commit:
- firefly/access_token sops slot for the importer's Firefly III API auth
(FIREFLY_III_ACCESS_TOKEN_FILE — was the missing piece causing 401s
from the API after the autoupload secret authenticated)
- nginx fastcgi_read_timeout=600s on the importer vhost (prevents 504
while PHP-FPM is still processing the batch)
- PHP-FPM max_execution_time=600s + memory_limit=512M on the importer
pool (PHP's stock 30s aborts mid-import for batches > ~50 transactions)
- timer re-enabled, wantedBy=[timers.target]
Caveats baked into a code comment:
- Sparda online-banking PIN must be [A-Za-z0-9] only. aqbanking 6.8.2's
-P pinfile mangles `:`, `+`, `'`, `?`, `@`, `%`, `*`; bank locks the
access (3 soft / 9 hard strikes) on rejected attempts. Same applies
whenever the sops secret is rotated.
- Bulk historical imports beyond the PSD2 90-day window need interactive
SCA approval per ~30-day chunk and cannot run from the timer; the
daily 35-day rolling window stays inside the no-SCA region.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
End-to-end FinTS pipeline against Sparda Südwest is wired up but
disabled — aqbanking 6.8.2's `-P pinfile` flag does not consume the
file content correctly on this build (verified: pinfile bytes match
the manually-typed PIN exactly, yet the bank receives a wrong PIN).
Three rejected attempts locked the access at Sparda; do not re-arm
the timer until the auth path is replaced (likely python-fints).
What works:
- aqbanking config and FinTS dialog (manual PIN entry)
- getaccsepa workaround for HKCAZ "Mussfeld 9160" rejection
- custom CSV profile (decimal amounts + IBAN columns) wired in
- Firefly importer auto-upload settings + sops secret slot
- inbox + profile-symlink tmpfiles
What's broken:
- Headless PIN delivery via aqbanking-cli -P
- Timer left wantedBy=[] so it cannot fire post-deploy
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Add a `metacfg.tools.wezterm` home-manager module so wezterm.lua
configuration can be reused across hosts instead of being duplicated
inline. Migrate halo and amd to the new module and enable it on rialo
(font size 14, term = xterm-256color).
Lays the groundwork for Sparda-Bank Südwest transaction sync via
direct FinTS (no third-party data proxy). aqbanking-cli in the system
PATH, persistent state at /var/lib/firefly-aqbanking, sops slot for
the online-banking PIN. Initial enrollment must be done interactively
on the host; systemd timer for automated fetches comes in a follow-up.
Share the check script via a parameterized mkDiskCheck function over
{ name, mountPoint, label } and iterate an attrset to emit the boot
and root services plus their daily timers.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
